Scaling Human-Object Interaction Recognition in the Video through Zero-Shot Learning

Table 3

The impact of the optical flow on the proposed system’s performance. The model trained with 1A + 2B and tested on 2A + 1B and all data.

Method	mAP (%)
	ALL data	Unseen data (2A + 1B)
2Stream – WE-VLAD	17.8	14.83
3Stream – WE-VLAD	19.21	16.65
2Stream + WE-VLAD	19.5	16.08
3Stream + WE-VLAD	20.84	17.32
2Stream + WE + VLAD (rnn)	20.86	16.96
3Stream + WE + VLAD (rnn)	21.27	17.63